Landscape drainage modelling to enhance Landsat classification accuracies

Studies integrating digital elevation models (DEMs) with multispectral digital satellite data have typically concentrated on geographic areas characterized by moderate to high topographic relief. Variables such as elevation, slope gradient and aspect contribute most significantly to the zonation of vegetation in these environments. In areas where relief is low, vegetation zonation is based not on individual form elements but rather on physical processes. The purpose of this research was to investigate the potential of integrating multispectral and ancillary process data in such a low relief environment. For this a study area was chosen in the Boreal forest of west central Alberta where the zonation of vegetation is based, to a large extent, on landscape drainage. An initial classification of forest cover based on Landsat multispectral data yielded overall classification accuracies of 58%. A DEM was developed from a digitized 1:50,000 topographic map sheet. The differential geometry of the DEM was mapped as a series of coverages: slope, aspect, and directional curvatures (down - and across slope). Two additional coverages, relief and flow paths, were also developed and mapped. A data set was extracted from the DEM through which landscape drainage could be evaluated. A univariate analysis of drainage using the form variables resulted in a 45% to 47% explanation of the observed variation. Multivariate analysis combining slope gradient, across and down slope curvatures, relief, and flow paths increased the explanation to 68%. The MSS data were reinterpreted integrating the DEM - based landscape drainage model. The resulting classification accuracy was increased to 73%.